Abstract

When solving a decision problem we want to determine an optimal policy for the decision variables of interest. A policy for a decision variable is in principle a function over its past; however, some of the past may be irrelevant and for both communicational as well as computational reasons it is important not to deal with redundant variables in the policies. In this paper we present a method to decompose a decision problem into a collection of smaller sub-problems s.t. a solution to the original decision problem can be found by solving the sub-problems independently. The method is based on an operational characterization of the future variables being relevant for a decision variable, thereby also providing a characterization of those parts of a decision problem that are relevant for a particular decision.